The Operations Bridge

The most expensive thing in your AI rollout isn’t the software. It’s the silence.

Here’s the failure mode nobody puts on a slide.

An operator tries the new AI tool. It hands back an answer that’s confident and wrong. They know it’s wrong — they’ve run this line for nine years. But saying so out loud means questioning the thing leadership just spent money on and staked their credibility to. So they don’t. They quietly stop trusting it, go back to the old way, and nod politely in the meeting where someone reports that adoption is going great.

You didn’t lose that rollout to a bad model. You lost it to a room where it wasn’t safe to say “this is wrong.”

MIT Technology Review put numbers on this in a December survey of 500 business leaders. 83% said psychological safety measurably improves the success of their AI initiatives. Eighty-three percent. And yet only 39% rated their own organization’s psychological safety as “very high.” Twenty-two percent admitted they’d hesitated to even lead an AI project — for fear of being blamed if it failed.

Read that gap again. Almost everyone knows culture decides whether AI works. Almost nobody is actually building the culture. As one CTO in the report put it: “psychological safety is mandatory in this new era of AI.”

On a plant floor, it isn’t just mandatory. It’s the whole game.

Because fear on the floor isn’t abstract. It’s not “career repercussions” in the HR sense. It’s specific, and it’s earned. Every operator I’ve worked with has watched an “efficiency” initiative show up with a smile and leave with a headcount number. So when AI arrives promising to make things faster, it doesn’t land on a blank page. It lands on a history. The tool says productivity; the floor hears layoffs.

You can have the best model in your industry. If the people who are supposed to use it believe it was built to replace them, they will not help you make it work. They’ll wait it out. And on the floor, waiting it out beats you every time — because the people closest to the work are the ones who decide whether the work actually changes.

This is where I part ways with how most companies run an AI program. They treat culture as the soft part — the change-management workstream you bolt on after the real work of picking a platform. That’s backwards. Culture isn’t the soft side of the rollout. It’s the operating system the rollout runs on.

Here’s the mechanism, concretely. AI is confidently wrong sometimes. That’s not a bug you’ll patch — it’s a permanent trait of the tool. The only defense is a human who checks its output against what they know to be true and says so out loud. That feedback loop — “the AI got this wrong, and here’s why” — is the single most valuable thing your people can hand you. And it’s exactly the thing fear kills first. In a low-trust plant, nobody reports the miss. They just lose faith in silence, and you end up flying blind on the one signal that would have made the tool better.

I think about adoption in three groups: the Builders who make things, the Users who run them, and the Resisters who won’t touch it. And the uncomfortable truth is that the Resisters are often the people closest to being replaced — the roles most exposed to automation resist the hardest, which is the most human thing in the world. Fear is what keeps a Resister a Resister. Psychological safety is the lever — maybe the only lever — that turns one into a User. Not another training. Not a slicker interface. Safety.

And it starts at the top of the room, not the bottom. That same MIT research found trust in AI climbed when leaders openly used it themselves — not when they mandated it, when they modeled it. On the floor that means a supervisor showing the team the prompt that flopped before the one that worked. It means a plant manager saying, “I asked the AI for this and it gave me garbage — here’s how I caught it.” Every time a leader admits the tool got it wrong and shows how they handled it, they’re not undermining the program. They’re building the exact muscle the program runs on.

On the floor, psychological safety for AI is boring and specific:

• Leaders share their own AI misfires — out loud, and first.
• The person who catches the AI being wrong gets thanked, not shushed.
• “I tried it and it didn’t work” is treated as data, not as failure.
• No AI-usage metric ever gets used to punish somebody.
• Nobody has to act on an answer they don’t understand.

None of that is a technology decision. All of it is a leadership one. And all of it is free.

So what? Your AI budget is buying capability you’ll never capture if your people are too afraid to tell you the truth about it. The model is the cheap part. Trust is the expensive part — and it’s the part that actually decides the outcome.

Now what? Before you approve the next tool, ask a harder question than “which platform.” Ask: on my floor, is it safe to say this AI is wrong? If you don’t know the answer, that’s your first project — and it matters more than anything in the vendor’s demo.

Culture eats your AI strategy for breakfast. Feed it on purpose.

Source: “Creating Psychological Safety in the AI Era,” MIT Technology Review Insights (in partnership with Infosys Topaz), December 2025, based on a survey of 500 business leaders.